Automating Exploratory Proteomics Research via Language Models

TL;DR

PROTEUS presents a fully automated, LLM-driven system for exploratory proteomics research that plans objectives, orchestrates specialized tools, iteratively refines analyses, and proposes data-grounded hypotheses from raw data. The hierarchical planning framework (objectives, workflows, steps) and dynamic feedback enable end-to-end analysis across CyTOF and clinical MS datasets, demonstrated by 191 hypotheses and multiple case studies. Evaluation combines automatic LLM scoring on five metrics with human expert reviews, showing reliable coherence and novelty with literature and providing insights into biological mechanisms. The work highlights a path toward AI-assisted, bias-m mitigating, scalable discovery in proteomics, while acknowledging limitations in knowledge elicitation and context management and pointing to extensions to multi-omics and broader biomedical research.

Abstract

With the development of artificial intelligence, its contribution to science is evolving from simulating a complex problem to automating entire research processes and producing novel discoveries. Achieving this advancement requires both specialized general models grounded in real-world scientific data and iterative, exploratory frameworks that mirror human scientific methodologies. In this paper, we present PROTEUS, a fully automated system for scientific discovery from raw proteomics data. PROTEUS uses large language models (LLMs) to perform hierarchical planning, execute specialized bioinformatics tools, and iteratively refine analysis workflows to generate high-quality scientific hypotheses. The system takes proteomics datasets as input and produces a comprehensive set of research objectives, analysis results, and novel biological hypotheses without human intervention. We evaluated PROTEUS on 12 proteomics datasets collected from various biological samples (e.g. immune cells, tumors) and different sample types (single-cell and bulk), generating 191 scientific hypotheses. These were assessed using both automatic LLM-based scoring on 5 metrics and detailed reviews from human experts. Results demonstrate that PROTEUS consistently produces reliable, logically coherent results that align well with existing literature while also proposing novel, evaluable hypotheses. The system's flexible architecture facilitates seamless integration of diverse analysis tools and adaptation to different proteomics data types. By automating complex proteomics analysis workflows and hypothesis generation, PROTEUS has the potential to considerably accelerate the pace of scientific discovery in proteomics research, enabling researchers to efficiently explore large-scale datasets and uncover biological insights.

Figures (11)

• Figure 1: (a) The framework of the iterative refinement of PROTEUS . (b) A detailed illustration of the working process of PROTEUS . First, the Data Description module generates a precise description of the proteomics dataset, based on which the Objective Planner proposes a series of research objectives. Based on each objective, the Workflow Planner then plans a sequence of analytical workflows, such as analyzing expression differences or labeling cell types. These planned workflows are executed using specialized bioinformatics tools and follow a fixed sequence of steps. The Workflow Updater and Objective Updater analyze the system's latest results, based on which they refine the subsequent workflows and objectives. PROTEUS produces numerous scientific hypotheses for each research objective, and continues its analysis until it reaches a pre-determined maximum number of objectives. These framework designs facilitate a robust, end-to-end proteomics research pipeline.
• Figure 2: Average scores and score distributions calculated on all 191 hypotheses, over 5 metrics.
• Figure 3: Score distributions on the 10 SPDB datasets, over 5 metrics. Specific dataset information corresponding to each index is provided in Table \ref{['tab:dataset-info']}.
• Figure 4: Summary statistics and workflow execution frequency
• Figure 4: Score distributions on the 2 clinical cohort datasets (HCC and GBM), over 5 metrics.